Reconstructions of past environments based on chemical "proxies" have provided invaluable insights into how the Earth system functions on a variety of timescales, from decades to millions of years. Organic molecular, or "biomarker", proxies are being applied increasingly in paleoclimate studies, and these compounds hold a wealth of information about an array of environmental conditions includin temperature, aridity, vegetation biome type, and water salinity. However, their current interpretation is mainly in terms of a single variable of interest, most often temperature.
This study, led by an early-career researcher from the Woods Hole Oceanographic Institution, is a first attempt to leverage the multidimensional quality of biomarker data for paleoclimate reconstruction. A statistical model will be developed using artificial neural network (ANN) and support vector regression (SVR) algorithms ("machine learning" techniques). This will allow quantitative evaluation of how assemblages of biomarkers in sedimentary archives relate to an array of modern environmental conditions. The outcomes of the proposed project will include development of a new toolbox to quantitatively reconstruct past climates over land and in the ocean in many types of paleoclimate archives. Such a tool could have broad application across a range of fields including environmental studies, petroleum geology, and computer science.
